When AI Runs the Pricing, Rivals May Quietly Learn to Overcharge You
Two companies' pricing bots never talk — but they can still end up acting like a cartel.
Imagine two rival gyms, two airlines, or two internet providers. Each one has an AI program that sets prices, learning by trial and error what makes the most money. Nobody programs the AI to be sneaky. It just experiments: raise the price a little, watch what happens to sales, adjust. Now imagine the same thing happens at every competitor at once.
This new paper from researchers Qixuan Zai and Randall Berry looks at that exact setup, with a twist: the thing being sold gets worse when it's crowded. Think of a mobile network that slows down when everyone streams at once, or a restaurant where a full house means slower service. Customers weigh two things — the price and how packed it feels — against how much they're willing to pay. The companies don't know that trade-off in advance. Their AI has to figure it out by trying things.
Here's the unsettling part. Earlier research showed that AI programs learning side by side can accidentally land on behaviors that look like price-fixing — everyone charges high, nobody undercuts, and no one ever sent an email about it. This paper shows the same drift can happen in crowded markets, where the pressure to be cheap is normally strong. The AI can learn to keep prices high and let congestion quietly do the work of limiting customers.
Why should you care? Because AI-set prices are already showing up in airlines, hotels, ride-hailing, and retail. If independent learning programs tend toward softer competition, your bills could creep up even when nobody breaks the law in the traditional sense. The authors suggest this matters for regulators too: the old rules about collusion assume humans talked. With AI, the collusion may never be spoken — just learned. The paper is theoretical, so it's a warning sign, not proof your next bill is rigged.
- AI pricing software learns by trial and error — and separate programs can drift into charging similarly high prices without any human coordination.
- The study adds crowding to the mix: services that get worse when busy, like slow networks or packed venues, where customers care about both price and congestion.
- It's a heads-up for regulators, since today's anti-collusion rules assume people talked to each other — AI may never need to.
Why It Matters
If AI sets prices everywhere, your bills may creep up even with no illegal handshake behind them.